Welcome to Voice for AI Agents and Applications, built in partnership with Vocal Bridge, an AI Fund portfolio company and taught by CEO Ashwyn Sharma. In this course, you'll learn how to build agents with voice interactive user interfaces. You add voice to your existing agents, as well as for agents that use voice as a function call tool. You also apply voice evals to your agents with detailed feedback on quality and accuracy to enhance them. The vast majority of people on this planet find speaking and listening much easier than writing and reading. And as voice UIs become more reliable, it'll open up many new applications for many people. For example, in The Batch newsletter, I wrote about my building a simple cat themed math quiz application for my seven-year-old daughter. She's enjoyed using the keyboard to play this game, but using Vocal Bridge, I was able to add a voice UI pretty quickly. So I can now quiz her verbally in a friendly way and she can respond verbally as well. And this really changes how the experience feels. Voice applications have historically faced a trade-off between latency and intelligence. One option is to use native voice-in, voice-out or voice-to-voice real-time models, and they're fast, but less reliable and harder to control. The other approach is to use a pipeline that inputs audio, uses speech-to-text, then has an LLM execute an agentic workflow, and then finally text-to-speech to read out the response. That's more reliable and more controllable, but adds latency, which is a problem for real-time conversations. In this course, you learn how to build voice applications that are both fast and reliable. Vocal Bridge, which you'll learn to use, relies on a custom architecture with a fast foreground agent for real-time conversation, as well as a background agent for reasoning workflows, guardrails, and tools. This gives your voice agents a best combination of both low latency and high intelligence. Thanks Andrew. This course teaches three integration patterns that meet you where you are. The first pattern is voice embedded in applications. Think of a game or a productivity tool where users can use voice commands, but they can also click and interact with the UI directly. The voice agent needs to trigger UI changes when the user speaks, and it needs to know when the user clicks on something. That bidirectional awareness is what makes the experience feel natural. The second pattern is voice for existing agents. You've already built a Claude, GPT, or LangChain powered agent with custom logic. You don't want to rewrite it. Vocal Bridge sits as a thin layer in front of your agent. It handles voice to intent conversion, but understanding when to delegate a question to your agent versus handling conversational pleasantries itself. The third pattern is the voice as a tool your LLM can call. Imagine a recruiting agent coordinating interviews. A candidate texts their availability and the agent responds in text to confirm and then also calls and coordinates the interview date that works the best for the candidate. Or a brainstorming agent, working with you via chat. It says, this would be faster to talk through, and opens a live voice session. So the agent chooses the right modality for the moment. We will also have Scott Johnston, former CEO of Docker, and board member of Vocal Bridge to share with us all the recent developments in bringing AI voice and the landscape of voice agents and applications to production. Many people have worked to create this course. I'd like to thank Eli Chen and Rakesh Utekar from AI Fund, and Aditi Dhar and Jitesh Gupta from Vocal Bridge. From DeepLearning.AI, Brendan Brown and Esmaeil Gargari also contributed to this course. In the first lesson, you will learn about the traditional voice agent stack in detail. So you understand what's being abstracted away by Vocal Bridge. Then you will see live demos of all three integration patterns we discussed. By the end of the lesson, you will know which pattern fits your use case and start building them in the next lessons. This sounds great. Let's get started.